基于深度学习的船舶舷号检测与识别
2021-05-23路云胡杰
路云 胡杰



摘要:陆地交通和水上交通是我国交通运输必不可少的两部分,而水上交通以船运为主。类似汽车车牌号,正规的船舶一般在两舷水线以上标明舷号以方便身份识别,但由于舷号命名和印刷的非标准性,其计算机的图像识别尚未进入实用阶段。本文基于EAST 的场景文本检测算法以及基于CRNN的端到端不定长文字识别算法,提出一种分阶段识别船舶舷号的解决方案。实验结果表明,该方案能较有效地对船舶舷号进行检测与识别,识别的准确率为73.06%。
关键词:深度学习;舷号检测;舷号识别;EAST;CRNN
中图分类号 TP391 文献标识码:A
文章编号:1009-3044(2021)11-0178-03
Detection and Identification of Ship's Hull Number Based on Deep Learning
LU Yun,HU Jie
(School of Computer Science, Yangtze University, Jingzhou 434000, China)
Abstract: Land transportation and water transportation are two essential parts of China's transportation, and water transportation is mainly based on shipping. Similar to car license plate numbers, regular ships generally have hull numbers above the waterline on both sides to facilitate identification. However, due to the non-standard naming and printing of hull numbers, computer image recognition has not yet entered the practical stage. Based on EAST's scene text detection algorithm and CRNN-based end-to-end variable length text recognition algorithm, this paper proposes a solution to identify hull number in stages. Experimental results show that this scheme can detect and identify ship's hull number more effectively, and the accuracy of identification is 73.06%.
Key words: deep learning; detection of hull number; identification of hull number; ESAT; CRNN
1 引言
水上交通的船運在陆地交通拥挤的情况下,以其耗能低、载量大和运费少的优点发挥着重要的交通运输作用,即使有港口调度,规定具体时间段的具体船舶通行时间及航道,但也没有发挥真正的优势,交通混乱、行驶不规范、撞击航标灯、船舶抢修不及时以及货船侧翻倾覆现象等时常发生。舷号像车牌号一样透露出该船的基本信息,例如中国海军小型舰艇和军辅船舷号命名规则如下:“1”字头为航母和驱逐舰,“2”或“3”字头为常规潜艇,“4”字头为核潜艇等[1]。运输船的舷号也有特定的命名规则,命名规则不统一,舷号位置不固定,印刷字体及大小不固定,要考虑到汉字和数字的多种组合。因此,港口调度结合船舶舷号的检测与识别可以得出该船舶的基本信息,方便事故追溯和及时抢修索赔,有利于规范行船行为,确保水上交通顺畅。……
